Magnetic Resonance Gibbs Artifact Removal And Parallel Imaging Based On Deep Learning

Deep learning technology has achieved remarkable results in the field of computer vision,especially in image recognition and detection,image restoration,and denoising.The Basic Convolutional Neural Network(CNN),and other variant networks based thereon,can automatically extract corresponding feature representations from a large amount of training data.Thus,complex structural information is identified,and a nonlinear mapping relationship between the input and the output can be constructed.Inspired by this technique,deep learning techniques can be used in some of the image processing techniques in Magnetic Resonance Imaging(MRI).One of the most influential patient experiences in magnetic resonance imaging is that the scanning time is too long.In the absence of sufficient sampled data,the resulting image will be disturbed by severe Gibbs artifacts.The imaging speed in magnetic resonance imaging technology depends on the number of samples,as well as the limitations of physiological and hardware conditions.Under these conditions,the under-sampling imaging technology of magnetic resonance can break through the sampling speed of magnetic resonance and utilize some algorithms.Techniques such as GRAPPA,SENSE,and Compressed Sensing Imaging can reconstruct images of the same quality as full sampling.This paper designs and adjusts the neural network based on deep residual network and U-NET to train artifacts with images to establish nonlinearity between images with artifacts and images without artifacts.Mapping relationships,while adding natural images to training,first through the natural image training network and then through the magnetic resonance image training,training can be more effective than the single data set to eliminate artifacts,and retain the structural information of the image,while research It is found that a sufficiently trained neural network can replace the data consistency unit and perform end-to-end image processing by itself.In this paper,the neural network based on deep residual network is designed and adjusted to train the parallel imaging mode of magnetic resonance imaging.The specially designed deep neural network is used to realize the non-sampled image between the input sampled image and the gold standard.The linear mapping relationship enables under-sampling reconstruction of magnetic resonance images.Compared with traditional methods,this method can process pictures more flexibly without setting additional parameters and achieving faster processing speed.The research in this paper shows that the deep convolutional neural network has the advantages of artifact elimination and under-sampling image reconstruction than traditional methods.Deep learning technology has great potential in areas such as magnetic resonance image processing.